BioMEMS-Based Coding for Secure MedicalDiagnostic Devices

Tuan Le, Gabriel Salles-Loustau, Laleh Najafizadeh, Mehdi Javanmard, Saman ZonouzElectrical and Computer Engineering Department, Rutgers University

Abstract—Trustworthy and usable point-of-care solutions re-quire not only effective disease diagnostic procedures to ensuredelivery of rapid and accurate outcomes, but also lightweightprivacy-preserving capabilities. In this paper, we present aBiomedical Microelectromachanical System (BioMEMS)-basedsensor for portable, inexpensive smartphone-based biomarkerdetection. The biosensor presented here provides the ability forsignal encryption at the physical sensor level to ensure patient’sdiagnostic confidentiality. Our results show that this designallow us to protect the samples measurements while accuratelydistinguish different test samples.

I. INTRODUCTION

Healthcare management and delivery costs in developedcountries are skyrocketing. In response to this trend, fed-eral agencies have supported diverse lines of applied re-search in the use of technology for health monitoring andintervention [5], [10]. Applications such as point-of-care(POC) diagnostic devices takes advantage of state-of-the-arttechnologies to compile information about medical healthsecurely and in real-time. POC enables a transition fromreactive, hospital-centered to preventive, patient-centered andcost-effective health care.

Flow cytometry in microfluidic has been studied exten-sively as an alternate method for diagnosing and monitoringdiseases such as HIV, malaria, and tuberculosis [1], [8].However, current techniques require substantial laboratorywork with stringent protocols. The possibility of integratingPOC systems with mobile platforms has also been veryrecently demonstrated through the diagnosis of a seriesof conditions including vitamin-D deficiency and KaposisSarcoma disease [4], [6], [7]. These solutions either storethe results in text file or distribute them over the network,exposing patient diagnostics to disclosure in case of databreach.

In this work we consider the design of a POC devicebiosensor for impedance cytometry. Our biosensor designaddresses three major requirements:

i) portability and low cost. Portable cytometry basedPOC devices can be used as the standard apparatus formedical diagnosis. The disposable biosensor contributes tothe low cost of the test procedure. The low-cost constraintmotivates the employment of widely deployed platformssuch as smartphones to provide computational resources. Thediagnostic solution should be easy-to-use to replace goldstandard laboratory methods where highly trained techniciansare required and must followed strict protocols.

ii) accuracy and performance. Due to the importanceof the medical diagnostic outcome, e.g., HIV diagnostic,the low-cost, portable biosensor should be designed withoutsacrificing the sensitivity of the diagnosis. The biosensorshould be able to differentiate the biomarkers for diagnosticpurposes. Furthermore, patients prefer accelerated testingprocedures over the traditional clinical visit with longer timereturn of test result. Therefore, by using impedance cytom-etry biosensor and signal analysis, the POC micorfluidicdevice can perform diagnosis and return the result quicklycomparing to the protracted practice at clinical visit.

iii) usable security and diagnostic confidentiality guar-antees. Patients are concerned about the confidentiality oftheir medical records [3]. Diagnostic disclosures may leadto undesired consequences such as insurance premium raisesand negative social impacts. Currently, medical institutionsare responsible for protecting the confidentiality of the pa-tient diagnostics. In the proposed solution, the encryptionmechanism is embedded in the sensor and is part of thedata acquisition process. The encryption key is stored in thesensor and remain in the possession of the user.

Current methods to protect medical data leverage generalpurpose digital encryption algorithms such a DES, AES,Blowfish. All these ciphers operate on digital data, gener-ating a digital ciphertext from a digital cleartext. In thispaper we introduce an analog signal encryption scheme thatgenerate an analog signal ciphertext. Contrary to scramblingtechniques which typically operate on a plaintext signal,our sensor encryption mechanism is embedded in the signalacquisition process. Similarly to digital encryption tech-niques, our cipher robustness relies on the confidentiality of arandomly generated encryption key. The key corresponds toa one-time pad: the values embedded in the key correspondto the sensor parameters during data acquisition. Without theknowledge of these parameters it is impossible to analyze theacquired signal.

This paper is structured as follow: Section II present ourapproach for cytometry and encryption, Section III detailsthe implementation the biosensor and Section IV show ourresults for our hardware-based encryption and test accuracy.

II. OVERVIEW

We introduce a new biosensor design for portable POCmicrofluidic diagnostic device with the ability of protect-ing the diagnostic confidentiality. The impedance cytometrybiosensor utilizes the hardware-based analog encryption to

1 4 3 2 . . . N

1 3 2 . . . N

(a)

(b)

(c)

Fig. 1. Design of the integrated system. (a) Model of operation. Input electrodes commonly connected to an AC voltage source. Output electrodes areconnected to an analog switch controlled by microprocessor. The controller randomly activates different subsets of electrodes, resulting in multiple peaksfor each blood cell detected. (b) Design of the biosensor with 2 and 3 independent output electrodes. (c) Design of the biosensor with 5 and 9 independentoutput electrodes.

protect the cytometry measurements of cells passing throughthe channel. The design of multiple independent outputscan dynamically change its configuration (i.e., active outputelectrodes) to obfuscate the impedance measurement beforesending it to the smartphone for signal analysis.

A microcontroller selects the random order of indepen-dent output electrodes and changes the configuration ofbiosensor active electrodes. The random selection of themicrocontroller is stored on the POC device as the encryptionkey. The encrypted data is sent to smartphone for peaksanalysis. The smartphone app analyzes the encrypted signaland counts the peaks caused by the impedance changesper passing particle through microfluidic device. The peaknumber does not necessarily correspond to the true number ofparticles that were presented, because more than one outputelectrodes may have been activated during data acquisition.The encryption key stored on the POC device corresponds tothe number of active electrodes. Thus, the app must send thecounted number of peaks back to the microfluidic device fordecoding. The POC microfluidic device recovers the numberof particles of different types and determines the user’sdisease condition through a simple threshold comparison, andnotifies the user accordingly.

III. SYSTEM DESIGN

Our design expands the simple impedance cytometer [9]but uses multiple electrodes with multiple inputs shortedtogether and multiple independent outputs. Fig. 1a describesthe operation model of the integrated system. We de-signed the microfluidic channel with an interdigitated multi-electrode pair configuration to mask the true count of cellspassing through the microfluidic channel. Fig. 1b and Fig. 1cshow the details of the computer aided design (CAD) ofthe sensing regions of the biosensors. The lead electrodein the array of output electrodes is defined as the electrodecomplemented only by one input electrode on one side. Thus,it will respond with a single voltage drop per passing cell;whereas the remaining output electrodes are surrounding onboth sides by common excitation electrodes. Each of theremaining electrodes in the sensing regions will respond witha signature of double peak per passing cell.

Multi-Electrode Signal Encryption: A cell passingthrough the sensor channel generates multiple peaks thanksto the multiple electrodes. The output of the electrodes areselected or discarded through the multiplexer using a pseudo-random selection. This selection is randomly generated bythe sensor microcontroller and constitutes the encryptionkey. The individual output from the electrodes are addedtogether to form the cipher text. This encryption approachprevents a potential eavesdropper, without access to thesignal encryption key, to discern the true number of cells thathave passed by the channel. The strength of the biosensor’ssignal encryption methodology relies on the biosensor’sreconfigurability to generate various signal fingerprints. Theencryption key corresponds to a one-time pad scheme. Thedata acquisition time frame is split in periods of variableduration (from a few seconds up to a min). For each ofthis period, a different pattern of electrode is scheduled tobe used. The key embeds the information about the periodlength and the sequence of electrode used.

System Integration: The microfluidic channel flow isdriven by the external peristaltic pump, i.e., Harvard Ap-paratus 11 Pico Plus Elite. The Raspberry Pi microcontrolleris used to generate the random selection sequence of theoutput electrodes in the microfluidic device through the 16:2multiplexer MAX14661. The selected output sequence of thesignal is recovered by the lock-in amplifier. We used a ZurichInstruments HF2IS impedance spectroscope coupled with aHF2TA trans-impedance amplifier to measure the electricalimpedance across the microfluidic channel. The input elec-trode of the microfluidic channel is excited with a combi-nation of [500, 800, 1000, 1200, 1400, 2000, 3000, 4000]kHzsignals. Excitation voltage is at 1V per excitation signal.The signal is sampled at 450Hz. The low pass filter is setto have cut off frequency at 120Hz.

To upload the encrypted signal to the signal analysis appon the smartphone, the controller (Raspberry Pi) is connectedto the smartphone through a micro-USB to the USB cable.The signal analysis app leverages the Android USB accessoryAPI [2], which allows the phone to detect the controlleras soon as it is connected and launches the diagnostic userinterface. The app has two purposes: it performs the signal

(a) (b)

(c) (d)Fig. 2. Representative encrypted cytometry data of a sensor with 9 input electrodes and 9 independent output electrodes detecting a single bead. Pseudo-random sequence selection of output electrodes. (a) Signal of single bead when electrode 9 selected. (b) Signal of single bead when electrode 1 and 9selected. (c) Signal of single bead when electrode 1,2, and 9 selected. (d) Signal of single bead when all electrodes selected (1-9). True number of peakscan only be detected/decrypted using unique key sequence.

Estimated bead counts - 7.80 um-50 0 50 100 150 200 250 300 350Em

piric

al b

ead

coun

ts

0

100

200

300

400

(a) 7.8 µm synthetic beads.

Estimated bead counts - 3.58 um-200 0 200 400 600 800 1000 1200Em

piric

al b

ead

coun

ts

0

200

400

600

800

(b) 3.58 µm synthetic beads.

Fig. 3. Measured bead count vs number of beads expected for differentconcentrations

analysis on the encrypted signal and it provides an interfacefor the user to start the blood test and provides a testprogression feedback to the user on the screen.

IV. EVALUATIONS

In our experiments, we evaluated the performance of thebiosensor using micron-sized synthetic beads (7.8 µm and3.58 µm - MicroChem) as well as blood cells, suspended inPBS 0.9%. The solution is driven through the microfluidicchannel at a rate of 0.08 µL.

To validate the flow rate in the microfluidic channel,the estimated fluid volume passing through the channelper unit of time is calculated using the time response ofthe bead passing an electrode pair. In Fig. 2, the estimate

response time for the single peak is 20ms. The approximateddistance each bead travels when passing through a pair ofelectrodes is 45 µm (25 µm pitch, and 20 µm of two halvesof electrode). The microfluidic channels are 30 µm in widthand 20 µm in height. By dividing the volume of the solutionpassing through a pair of electrodes in the channel at theapproximated time, the actual flow rate in the channel canbe calculated to be 0.081 µL/min.

Fig. 2 illustrates how the sensor duplicates a peak gener-ated for one electrode into multiple peaks signals to preventthe disclosure of number of beads passing through thechannel. The figure shows the response of the biosensor tothe 7.8 µm synthetic bead solution at 2MHz. Fig. 2a showsthe measured response of the biosensor when the lead outputelectrode is selected and the remaining output electrodes arerouted to the ground port. Fig. 2b shows the response wherethe lead electrode is selected along with the last electrode.The lead electrode responds with a single peak whereas theremaining electrodes respond with double peaks per passingbead. Fig. 2c shows the response of the biosensor when thelead electrode 9 and electrode 1, 2 are selected. Fig. 2d showsthe outcomes when all the electrodes are activated.

In peak-analysis, the accuracy of the biosensor is evalu-ated by comparing the empirically detected peaks and theestimated number of synthetic beads passing through themicrofluidic channel. We diluted the 7.8 µm and 3.58 µmsynthetic beads with PBS, which is a commonly used bio-logical buffer that mimics physiological samples like blood.Synthetic beads are diluted at different concentrations toevaluate the empirical peak detection. The estimated numberof elements in the solution is calculated according to the con-centration information provided by the manufacturer. Foursamples of each concentration are collected. The bead countdata is taken from the first 5min from each sample. Fig. 3aand Fig. 3b show the correlation of the empirical peak detec-

Amplitude (V) - 500kHz0 0.005 0.01 0.015 0.02

Ampl

itude

(V) -

250

0kH

z#10-3

-1

0

1

2

3

4

5

6

7

8

3.58um Beads7.8um BeadsBlood Cells

Fig. 4. Impedance differences of synthetic beads and blood cells.

0.11 0.2150.3430.452

0.81

1.554

0

0.5

1

1.5

2

240607 481214 962428

Proc

essi

ng ti

me

(s)

Sample size

Computer - Intel i7-4710MQ (16GB RAM)

Nexus 5 - Qualcomm MSM8974 Snapdragon 800(2GB RAM)

Fig. 5. Performances of data analysis on a computer and smartphone

tion to the estimated peak counts in the microfluidic channelfor 7.8 µm and 3.58 µm synthetic beads. As expected, theempirical peak detection correlates to the estimated peakswhen varying the concentrations. The discrepancy in beadcounts is due to several reasons. For synthetic beads, thelonger the duration of the experiments, the more error wouldbe expected in the empirical bead counts as many beadssediment to the bottom of the inlet well never making itdownstream to the sensor in the micro-channel. The otherreason for the bead count loss is due to the beads beingadsorbed to microfluidic channel walls. These are issuesthat can be ultimately resolved with optimization of channelmaterial and surface chemistry, which was beyond the scopeof the current work.

To validate the accuracy of the biosensor, we performedruntime diagnosis analysis multiple times over several bloodsamples. The typical diagnostics procedure takes a 0.1mLof blood sample and completes all steps, including sen-sor encryption, mobile app signal processing, peak countsdecoding and diagnostics, within 1 minute. Fig. 4 showsthe impedance differentiation between the blood cells andsynthetic beads at different excitation frequencies. Thus, thebiosensor can utilizes the difference of dielectric propertiesand concentrations of cell types to give a reliable mobilediagnosis.

Fig. 5 shows performance comparison of the peak detec-tion algorithm, when it runs on a standard computer system(possibly a cloud virtual machine) and on a smartphone

device. It is noteworthy that a standard system providesmuch better performance than a mobile device, as the sam-ple size grows larger. Aside from the storage capabilities,the enhanced computing power motivates the use of acloud based service for handling peak detection and post-processing rather than using the smartphone. For smallersamples, however, the smartphone’s app could be configuredto perform the peak counting signal processing locally.

V. CONCLUSIONS AND FUTURE WORKS

In this paper, we introduced a new application forBioMEMS-based sensing in portable POC diagnostics solu-tions that provide secure, low-cost, and accurate outcomesthrough the use of smartphone computational resources.We described a in-sensor hardware-based analog signal en-cryption that enables cloud-based analysis for encryptedanalog signals without disclosing measurements values. Weimplemented and integrated the biosensor circuitry and thesoftware stack and evaluated its accuracy empirically usingred blood cells and multiple solutions of synthetic beads.

The biosensor enables data encryption of the cell count atthe hardware level. However, the encrypted signal still carriesinformation about the cells. Specifically, the amplitude orwidth of the response peak can reveal information aboutthe composition or shape of the cell. As future work, weplan to leverage two more parameters to protect both ofthese information: application of random gains to each outputelectrode to modify the original amplitude, and also contin-uous alteration of the fluid speed to create arbitrary widthsfor passing cells of identical types. These two parameterscan be incorporated as part of the encryption key for signaltransformation.

REFERENCES

[1] P. Balakrishnan, M. Dunne, N. Kumarasamy, S. Crowe, G. Subbu-lakshmi, A. K. Ganesh, A. J. Cecelia, P. Roth, K. H. Mayer, S. P.Thyagarajan, and S. Solomon. An inexpensive, simple, and manualmethod of cd4 t-cell quantitation in hiv-infected individuals for use indeveloping countries. JAIDS Journal of Acquired Immune Deficiency

Syndromes, 36(5):1006–1010, 2004.[2] A. Developers. USB Accessory. https://developer.android.com/guide/

topics/connectivity/usb/accessory.html, 2015. [Online; accessed 19-July-2015].

[3] J. M. Eisenberg. Can you keep a secret? Journal of general internal

medicine, 16(2):131–133, 2001.[4] S. Lee, V. Oncescu, M. Mancuso, S. Mehta, and D. Erickson. A

smartphone platform for the quantification of vitamin d levels. Lab

on a Chip, 14(8):1437–1442, 2014.[5] C. LeRouge, V. Mantzana, and E. V. Wilson. Healthcare information

systems research, revelations and visions. European Journal of

Information Systems, 16(6):669, 2007.[6] X. Liu, T.-Y. Lin, and P. B. Lillehoj. Smartphones for cell

and biomolecular detection. Annals of biomedical engineering,42(11):2205–2217, 2014.

[7] M. Mancuso, E. Cesarman, and D. Erickson. Detection of kaposi’ssarcoma associated herpesvirus nucleic acids using a smartphoneaccessory. Lab on a Chip, 14(19):3809–3816, 2014.

[8] F. E. McKenzie, W. A. Prudhomme, A. J. Magill, J. R. Forney,B. Permpanich, C. Lucas, R. A. Gasser, and C. Wongsrichanalai.White blood cell counts and malaria. Journal of Infectious Diseases,192(2):323–330, 2005.

[9] J. Mok, M. N. Mindrinos, R. W. Davis, and M. Javanmard. Digitalmicrofluidic assay for protein detection. Proceedings of the National

Academy of Sciences, 111(6):2110–2115, 2014.[10] P. Shekelle, S. C. Morton, and E. B. Keeler. Costs and benefits of

health information technology. 2006.